[51Nod 1227] 平均最小公倍数 (杜教筛)

题目描述

求$largesum_{i=a}^bsum_{j=1}^ifrac{lcm(i,j)}i$
$1 <= a <= b <= 10^9$

题目分析

这道题其实是[51Nod 1238] 最小公倍数之和题解的简化版,或者说是本质…
就直接上公式了

令$large f(n)=sum_{i=1}^nfrac{lcm(n,i)}n=sum_{i=1}^nfrac i{(n,i)}$,则$Ans=sum_{i=a}^bf(i)$
$large f(n)=sum_{d|n}sum_{d|i}frac id[(i,n)==d]\=sum_{d|n}sum_{d|i}frac id[(frac id,frac nd)==1]\=sum_{d|n}sum_{i=1}^{lfloorfrac nd floor}i[(i,frac nd)==1]\=sum_{d|n}sum_{i=1}^di[(i,d)==1]$
此处有一个常识
$sum_{i=1}^ni[(i,n)==1]=frac {varphi(n)n+[n==1]}2$

• 证明如下
  • 当$n>1$时,若$(i,n)=1iff(n-i,n)=1$,所以与$n$互质的数是成对出现,且他们的和为$n$
  • 再加之$n=1$的特殊情况,可得
    $sum_{i=1}^ni[(i,n)==1]=frac {varphi(n)n+[n==1]}2$

继续
$large herefore f(n)=sum_{d|n}frac {varphi(d)d+[d==1]}2\=frac{1+sum_{d|n}varphi(d)d}2$
$large herefore sum_{i=1}^nf(i)=frac{n+sum_{i=1}^nsum_{d|n}varphi(d)d}2\=\frac{n+sum_{d=1}^nvarphi(d)dsum_{d|i}1}2\=frac{n+sum_{d=1}^nvarphi(d)dlfloorfrac nd floor}2$
此时就可以用整除分块优化+杜教筛计算$large sum_{d=1}^nvarphi(d)d$
令$large h(n)=varphi(d)d,g(n)=sum_{i=1}^nh(d)$
$largeecause n=sum_{d|n}varphi(d)\ herefore n^2=sum_{d|n}varphi(d)n=sum_{d|n}varphi(d)dcdotfrac nd=sum_{d|n}h(d)frac nd\ herefore sum_{i=1}^ni^2=sum_{i=1}^nsum_{d|i}h(d)frac id\=sum_{d=1}^nh(d)sum_{d|i}frac id\=sum_{d=1}^nh(d)sum_{i=1}^{lfloorfrac nd floor}i$
注意我们是在杜教筛,不能到这里就把$sum_{i=1}^{lfloorfrac nd floor}i$看做$Theta(1)$可求的式子而之后再也不做变换,那样往往会陷入更麻烦的方法或者死胡同里去,接着往下
$large herefore sum_{i=1}^ni^2=sum_{i=1}^nisum_{d=1}^{lfloorfrac ni floor}h(d)=sum_{i=1}^nicdot g({lfloorfrac ni floor})\ herefore g(n)= sum_{i=1}^ni^2-sum_{i=2}^nicdot g({lfloorfrac ni floor})$
然后套杜教筛即可

虽然在外面套了一层整除分块优化,但由于记忆化的原因,不影响时间复杂度,预处理出一部分$g$后总复杂度为$largeTheta(n^{frac 23})$

…有兴趣的可以去了解一下[51Nod 1238] 最小公倍数之和题解,比这道题恶心点

AC code

#include <cstdio>
#include <cstring>
#include <cmath>
#include <tr1/unordered_map>
#include <algorithm>
using namespace std;
using namespace tr1;
typedef long long LL;
const int inv6 = 166666668;
const int inv2 = 500000004;
const int MAXN = 5e6 + 1;
const int mod = 1e9 + 7;
int Prime[MAXN], phi[MAXN], Cnt;
bool IsnotPrime[MAXN];
LL g[MAXN];
void init()
{
	phi[1] = 1;
	for(int i = 2; i < MAXN; ++i)
	{
		if(!IsnotPrime[i])
			Prime[++Cnt] = i, phi[i] = i-1;
		for(int j = 1; j <= Cnt && i * Prime[j] < MAXN; ++j)
		{
			IsnotPrime[i * Prime[j]] = 1;
			if(i % Prime[j] == 0) { phi[i * Prime[j]] = phi[i] * Prime[j]; break; }
			phi[i * Prime[j]] = phi[i] * phi[Prime[j]];
		}
	}
	for(int i = 1; i < MAXN; ++i)
		g[i] = (g[i-1] + 1ll * phi[i] * i % mod) % mod;
}
unordered_map<LL, LL> G;
inline LL sum1(LL i, LL j) { return ((i+j)%mod) * ((j-i+1)%mod) % mod * inv2 % mod; }
inline LL sum(LL n)
{
	if(n < MAXN) return g[n];
	if(G.count(n)) return G[n];
	LL ret = (n%mod) * ((n+1)%mod) % mod * ((2*n+1)%mod) % mod * inv6 % mod;
	for(LL i = 2, j; i <= n; i=j+1)
	{
		j = n/(n/i);
		ret = (ret - sum(n/i) * sum1(i, j) % mod) % mod;
	}
	return G[n]=ret;
}

LL solve(LL n)
{
	LL ret = n%mod;
	for(LL i = 1, j; i <= n; i=j+1)
	{
		j = n/(n/i);
		ret = (ret + ((sum(j)-sum(i-1))%mod) * ((n/i)%mod) % mod) % mod;
	}
	return ret * inv2 % mod;
}

int main ()
{
	LL a, b; init();
	scanf("%lld%lld", &a, &b);
	printf("%lld
", ((solve(b)-solve(a-1)) % mod + mod) % mod);
}